Using master fader for more headroom?

[ScuzzyEye]

And just so no one can call me out, the mantissa is actually 23-bits, but can be positive or negative, as indicated by the sign bit. So for the purposes of audio samples that represent alternating current, you effectively have 24-bits.

[odarmonix]

Doesn't happen with floating point. You always have 24-bits of precision in the mantissa, the exponent changes as the scale of the values change. Floating point is very well suited for PCM encoding. The size of the steps between values get smaller as the values themselves get smaller.

A crude example would be values from 0.1 to 1 get 10 steps:
0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9, 1.0

When you go up a level magnitude from 1 to 10, you still get 10 steps:
1, 2, 3, 4, 5, 6, 7, 8, 9, 10

And one more, from 10 to 100:
10, 20, 30, 40, 50, 60, 70, 80, 90, 100.

Except you actually get 16 million steps at each scale level, and there are 256 scales.

I'm well aware of the purpose of floating point audio and that it basically offers virtually unlimited headroom. But I'm not sure this is truly relevant here, how many musicians use floating point audio when sending their tracks to mastering studios ? Us Reason users have floating point audio processing for certain things like the SSL, but we're still limited to a maximum of fixed 24 bit for import & export.

That being said, does anyone REALLY need the insane dynamic range allowed by 32 bit float (let alone 64) for PCM encoding ? While the theory sounds good on paper it sounds like complete overkill to me, I'm not even sure it's all that useful for DAWs' internal processing and I don't remember having ever needed more than fixed 24 bit for any kind of task, although I can definitely hear the difference between 16 and 24 in some very specific situations.

Regardless of the above, in my previous post I only wanted to point out that working at extremely weak levels can be just as a bad idea as working with too hot levels, even though irreversible signal degradation is less likely with the former. Music definitely has been WAY more often butchered by excessively hot levels during the last two decades.

[ScuzzyEye]

I'm well aware of the purpose of floating point audio and that it basically offers virtually unlimited headroom. But I'm not sure this is truly relevant here, how many musicians use floating point audio when sending their tracks to mastering studios ? Us Reason users have floating point audio processing for certain things like the SSL, but we're still limited to a maximum of fixed 24 bit for import & export.

[...]

Regardless of the above, in my previous post I only wanted to point out that working at extremely weak levels can be just as a bad idea as working with too hot levels, even though irreversible signal degradation is less likely with the former. Music definitely has been WAY more often butchered by excessively hot levels during the last two decades.

I'm not even talking about the dynamic range of 32-bit floating point, but specifically another property offered by using floating point values for PCM samples. Because of how floating point works on computer platforms when you turn the volume done you don't lose steps between quantization levels. As the values stored become smaller the steps between them also get smaller. So you still have the same number of quantization levels, and lower volumes don't increase the SNR.

Reason 7 gained 32-bit import, so we've had that for a while. But alas, still no 32-bit export.

Additionally no one was talking about extremely weak levels. The mastering engineer was asking for an output that was at -6 dB. That's one bit. Even if the level is turned down 30 dB, that's still only 5 bits, and with 24 bit audio, a that maintains an SNR of 114 dB. Beyond the human threshold of detection, and nearly every DAC to even reproduce.

[odarmonix]

I'm not even talking about the dynamic range of 32-bit floating point, but specifically another property offered by using floating point values for PCM samples. Because of how floating point works on computer platforms when you turn the volume done you don't lose steps between quantization levels. As the values stored become smaller the steps between them also get smaller. So you still have the same number of quantization levels, and lower volumes don't increase the SNR.

Reason 7 gained 32-bit import, so we've had that for a while. But alas, still no 32-bit export.

Additionally no one was talking about extremely weak levels. The mastering engineer was asking for an output that was at -6 dB. That's one bit. Even if the level is turned down 30 dB, that's still only 5 bits, and with 24 bit audio, a that maintains an SNR of 114 dB. Beyond the human threshold of detection, and nearly every DAC to even reproduce.

It indeed can import 32 bit floating point audio, I mistakenly assumed it still was limited to 24 bit as importing a fixed 32 bit file didn't work on my end. I still don't see the usefulness in going above a 24 bit resolution for simple PCM files however ... at least not in a musical context.

No extremely weak levels may have been evocated and one bit out of 24 obviously isn't a huge loss. It's only this part that I strongly disagree with :

that means you can lower your mix as much as you want

"As much as you want" sounds to me like an individual could decide to lower his levels down to -200 dB or even less just for the sake of avoiding clipping before exporting, without the fear of any kind of side-effect, simply because he/she is unaware of some digital audio rules. Even if this will most likely have no dramatic consequences with a really high bit depth, it just isn't the right thing to do and those who export at lower bit depths will run into trouble by doing this, that's all I'm trying to say.

[ScuzzyEye]

It indeed can import 32 bit floating point audio, I mistakenly assumed it still was limited to 24 bit as importing a fixed 32 bit file didn't work on my end. I still don't see the usefulness in going above a 24 bit resolution for simple PCM files however ... at least not in a musical context.

"As much as you want" sounds to me like an individual could decide to lower his levels down to -200 dB or even less just for the sake of avoiding clipping before exporting, without the fear of any kind of side-effect, simply because he/she is unaware of some digital audio rules. Even if this will most likely have no dramatic consequences with a really high bit depth, it just isn't the right thing to do and those who export at lower bit depths will run into trouble by doing this, that's all I'm trying to say.

At its worst 32-bit floating point hits only 144 dB SNR. The same as 24-bit audio. In other cases it can do better, but you're only absolutely guaranteed the same quantization noise level as 24-bit integer. The reason to use floating audio is that it can't clip. So you can export your stems, or your stereo mix for mastering, and not have to worry one bit if you've left enough headroom. That's why I'll be excited if Reason ever gets 32-bit float export.

With 32-bit floating export you could turn down the master to -200 dB, and then re-import and turn it back up, and likely the same signal, save for a couple rounding errors at the 10th decimal place. But yeah, 24-bit wouldn't survive that. And "as much as you want," could be taken to mean -200, but who'd really want to do that?

[MAL9000]

And this is why I love this place. Very interesting guys. Please keep the discussion going as I am learning LOADS

[odarmonix]

At its worst 32-bit floating point hits only 144 dB SNR. The same as 24-bit audio. In other cases it can do better, but you're only absolutely guaranteed the same quantization noise level as 24-bit integer. The reason to use floating audio is that it can't clip. So you can export your stems, or your stereo mix for mastering, and not have to worry one bit if you've left enough headroom. That's why I'll be excited if Reason ever gets 32-bit float export.

With 32-bit floating export you could turn down the master to -200 dB, and then re-import and turn it back up, and likely the same signal, save for a couple rounding errors at the 10th decimal place. But yeah, 24-bit wouldn't survive that. And "as much as you want," could be taken to mean -200, but who'd really want to do that?

If floating point basically prevents both clipping and quantization errors, then why has this thread been created in the first place ? The fact some people worry so much about clipping before sending off their tracks to mixing/mastering pretty much means to me that they don't export at floating point bit depths, either because they don't know what its real benefits are or because they just don't see the point of it. Again, if clipping is such an issue for them, they shouldn't ignore its opposite side effect.

-200 dB sure is an extreme example, but let's imagine this with more reasonable values : an average peak level of -48 dB at 16 bit already does some serious damage ! This looks like a ridiculously obvious thing to experienced tech-savvy people, but not every computer musician knows about the ins and out of digital audio. I know I've made this kind of mistake back when I had no clue how all of this works and I can tell you I've met people in the past who exported and shared their tracks at 22 KHz 8 bit, just because they had no idea what they were doing (although some of them still complained about noise they couldn't get rid of).

I personally won't be excited at all if PH ever add 32 bit float export to Reason, they may as well include DSD64 support while they're at it.

[ScuzzyEye]

If floating point basically prevents both clipping and quantization errors, then why has this thread been created in the first place ? The fact some people worry so much about clipping before sending off their tracks to mixing/mastering pretty much means to me that they don't export at floating point bit depths, either because they don't know what its real benefits are or because they just don't see the point of it. Again, if clipping is such an issue for them, they shouldn't ignore its opposite side effect.

-200 dB sure is an extreme example, but let's imagine this with more reasonable values : an average peak level of -48 dB at 16 bit already does some serious damage ! This looks like a ridiculously obvious thing to experienced tech-savvy people, but not every computer musician knows about the ins and out of digital audio. I know I've made this kind of mistake back when I had no clue how all of this works and I can tell you I've met people in the past who exported and shared their tracks at 22 KHz 8 bit, just because they had no idea what they were doing (although some of them still complained about noise they couldn't get rid of).

I personally won't be excited at all if PH ever add 32 bit float export to Reason, they may as well include DSD64 support while they're at it.

Why? I'd guess because most mastering engineers aren't computer engineers, and don't completely understand what 32-bit floating point offers. Plus it wasn't until recently that most DAWs started supporting 32-bit float output. Plus you can't play back floating point audio, without first mapping it to 24-bit integers so it can be sent to the DAC. That involves adjusting the volume to make sure it doesn't clip at that point. If the mastering engineer doesn't want to have to deal with turning audio that peaks at -3 dB down 3 dB to make it -6, then they are not going to want to have to deal with audio that could possibly peak at +3 dB.

-48 dB on 16-bit effectively makes it 8-bit. But if you add dithering the SNR should still be acceptable for listening, just not pre-mastering work. I will admit after thinking about it, -200 dB may be an issue for 32-bit float. I'll do the math but there's likely to be 2 decimal places of truncation. It likely tops out at -192 dB of attenuation with no loss. 200 dB is probably one and half bits too much.

32-bit float export is extremely useful. Especially if you want to export stems, and start a new project for mixing. Your export will match exactly what was being sent to the mixer while composing. DSD is useless as a working format, it's only use is for distribution. The only thing you can do is adjust the volume. You can't EQ, filter, add effects, or anything that someone processing audio would want to do. It's also lossy, or at least there's no way to convert PCM to DSD back to PCM and have the same result in and out. The format was more about copy protection, and licensing fees for Sony than being a replacement for PCM.

[odarmonix]

Why? I'd guess because most mastering engineers aren't computer engineers, and don't completely understand what 32-bit floating point offers. Plus it wasn't until recently that most DAWs started supporting 32-bit float output. Plus you can't play back floating point audio, without first mapping it to 24-bit integers so it can be sent to the DAC. That involves adjusting the volume to make sure it doesn't clip at that point. If the mastering engineer doesn't want to have to deal with turning audio that peaks at -3 dB down 3 dB to make it -6, then they are not going to want to have to deal with audio that could possibly peak at +3 dB.

-48 dB on 16-bit effectively makes it 8-bit. But if you add dithering the SNR should still be acceptable for listening, just not pre-mastering work. I will admit after thinking about it, -200 dB may be an issue for 32-bit float. I'll do the math but there's likely to be 2 decimal places of truncation. It likely tops out at -192 dB of attenuation with no loss. 200 dB is probably one and half bits too much.

32-bit float export is extremely useful. Especially if you want to export stems, and start a new project for mixing. Your export will match exactly what was being sent to the mixer while composing. DSD is useless as a working format, it's only use is for distribution. The only thing you can do is adjust the volume. You can't EQ, filter, add effects, or anything that someone processing audio would want to do. It's also lossy, or at least there's no way to convert PCM to DSD back to PCM and have the same result in and out. The format was more about copy protection, and licensing fees for Sony than being a replacement for PCM.

There's another important thing you haven't mentionned : floating point audio files are considerably bigger than their fixed point counterpart. I can't help thinking this is essentially a waste of disk space for anyone who knows how to set levels correctly. 24 bit still remains the best compromise in my experience, even for stems.

I hope you're not serious when you say 8 bit should be acceptable in a listening context (even with properly applied dithering), except maybe for material with next to zero dynamics like chiptune music or an atrociously overcompressed album. 8 bit used to be tolerable in early 90s video games, back when audio files couldn't take too much space on the support and long before lossy compression formats such as mp3 were invented. Aside from deliberate sonic destruction or things that don't require high fidelity like toys, I honestly can't think of any useful application for such low bit depths in 2016.

And yes, DSD eventually turned out to be an utterly pointless format as well as a epic commercial failure. This is just an example amongst many others that more doesn't always equal better.

[ScuzzyEye]

Yeah, 32-bit files are 33% larger than 24-bit files. Neither one makes even the slightest impression on a 1 TB hard drive.

I'm serious, 48 dB of dynamic range is more than most every song I listen to (mostly modern electronic), so 8-bits is fine. Dithering hides the quantization noise. The more serious loss in 90s video games was the 22 kHz sampling rate. That's unacceptable.

[Stranger.]

ΣΣΣ

[normen]

8bit means exactly the same as 48dB signal to noise ratio, nothing more, nothing less. There is no "pixels" in audio. So unless you have important audio information at -48dB its fine, really.

[Stranger.]

ΣΣΣ

[normen]

PCM (Pulse Code Modulation) isn't a file format, its a way of digitizing continuous waveform data, most audio containers (i.e. WAV, AIFF, CAF, MP4 etc.) can store PCM data. Few programs just resample audio without asking the user if they should do so.

[Stranger.]

ΣΣΣ

[ScuzzyEye]

Pulse Code Modulation isn't a file format, its a way of digitizing continuous waveform data, most audio containers (i.e. WAV, AIFF, CAF, MP4 etc.) can store PCM data. Few programs just resample audio without asking the user if they should do so.

It's stored as a format normen-this type of answer makes me doubt your a paid 'pro.
Straight from Microsoft's own instructions "uncompressed PCM wave formats" **my computer properties summary reports wavs as that format.
I feel i understand my sources,and modulation,not sure if you do so far.

We're talking about PCM in the abstract. Not stored in files, or even being played or measured. Just simply the rules on how digitally sampled audio behaves. No channels either.

[normen]

It's stored as a format normen-this type of answer makes me doubt your a paid 'pro.
Straight from Microsoft's own instructions "uncompressed PCM wave formats" **my computer properties summary reports wavs as that format.
I feel i understand my sources,and modulation,not sure if you do so far.

You go quote microsoft when its about pro audio

[Stranger.]

ΣΣΣ

[ScuzzyEye]

Sample accuracy/synchronization

Only matters for recording and playback. Not what we're talking about here.

Also doesn't apply to digital synths operating in the box.

[normen]

Heres the part that should tell you about the difference of PCM per se and storing PCM data:

PCM audio is stored, in a manner directly compatible with the AES3 digital audio interface.

Go for your tin hat- i got mine.-- it's all about the crystals and components-lords and ladies!!

..as soon as you go to or from digital audio stored as data inside a computer system to digital or analog audio as an electrical (or optical) signal.

[Stranger.]

ΣΣΣ

[normen]

I'm sorry,but it does very much apply here with converting in different operations.

Whats "converting in different operations" supposed to mean

[Stranger.]

ΣΣΣ

[normen]

Normen-have'nt you got a NULL TEST to go fix? i put it here in the forum for you.

I feel like I just tested a null already xD

[ScuzzyEye]

I'm sorry,but it does very much apply here with converting in different operations.
I understand it's all symbology..

You can be as sorry as you want, but clocks only come into play when ADCs or DACs are involved. Audio stored, and modified by a computer is always perfectly in sync. Even sample rate conversion, maintains perfect clocking, even if it is in itself an imperfect process.